Neurotechnology specialists at the MIT Media Lab in Cambridge, Massachusetts, under the direction of French researcher Nataliya Kosmyna, conducted a study on 54 volunteers aged 18 to 39 from around a dozen countries to assess the effects of ChatGPT use on brain function.
The study design
The study involved students and postdoctoral researchers from the Massachusetts Institute of Technology (MIT), Harvard University, and Wellesley College. Participants were equipped with headsets designed to read brain activity, specifically measuring information flows through different brain regions. The volunteers were instructed to write essays on three separate occasions, either utilizing ChatGPT or writing without its assistance, and subsequently asked to recall their work. Four months later, 18 of these volunteers participated in a follow-up, switching their roles.
Initial results and limitations
The results of this experiment, published online on Arxiv though not yet peer-reviewed, indicated that individuals who exclusively used ChatGPT for writing experienced a decline in “neural, linguistic, and behavioral” performance after a four-month period. The scientists acknowledged the study’s limitations, including:
- A very small sample size.
- A lack of statistical representation.
Despite these caveats, they noted that the findings “raise concerns about the long-term educational implications of [reliance]” on AI tools.
Community reaction and controversy
The publication of the 206-page study generated considerable discussion. Evolving AI, a platform with three million followers, posted a message stating: “Breaking: MIT just completed the first brain scan study of ChatGPT users and the results are terrifying,” which included misleading images not from the study.
Conversely, French entrepreneur Etienne de Saint-Martin stated that “the study contains numerous methodological biases” and characterized it as “not science, this is an opinion piece.”
AI consultant Joseph D. Stec described experiencing “one of the most ironic moments in recent AI research,” observing that people rushed to summarize the study using ChatGPT. Kosmyna confirmed that “around 90% of the media coverage came from automatically generated summaries,” expressing regret regarding “a very strong confirmation bias.”
Dutch epidemiologist Jan van den Brand noted that the study went viral because it was “misunderstood, misinterpreted and distorted,” adding, “That would be my worst nightmare as a scientist. Please, read the full article or at least the section on limitations.”
AI made executives worse at stock picking
Specific findings and further discussion
The study specifically found that when participants utilized ChatGPT, their brain activity, particularly in regions associated with attention, planning, and memory, significantly decreased. Minutes after using the tool, participants struggled to recall what they had submitted. In terms of evaluation, essays generated by ChatGPT received favorable grades but were found to be more stereotypical and less nuanced.
Professor Daniel Russo from Aalborg University commented, asking if this “sound familiar?” He suggested the study raises “the uncomfortable question” of whether users genuinely feel they have engaged in thorough thought processes. Russo continued, “Software professionals might recognize this feeling. You copy a code snippet, paste it in, tweak one variable and move on. But did you really solve the problem?”
The MIT Media Lab team has since launched a website to facilitate further discussion, receiving proposals for scientific collaborations from around the world. Kosmyna reported, “This is about our brains and new tools to master. It’s good news that so many people care!”
